Image Matching Technology Based On Binocular Vision Research

Binocular stereo vision technology is an important research field of computervision, which is widely applied to the intelligent robot navigation, virtual three-dimensional reconstruction, remote sensing, etc. Stereo matching algorithm is akey technology for binocular stereo vision, which its accuracy will affect the resultof applications of stereo vision technology directly.Stereo matching obtains a dense disparity map by searching one-to-onecorrespondence between projection points in the projected images from differentviewpoints of the same actual scene. It is the core part of stereo vision. However,considering the influence of noise, occlusion, depth discontinuities, texturelessregions and other factors, stereo matching is still a challenging work to obtain adense disparity map with high accuracy.From the perspective of improving the matching accuracy, this paper focuseson the global and local optimization algorithm respectively. In the stereo matchingalgorithm based on global optimization, we proposed two improved methodsaccording to the disadvantage of tradition belief propagation algorithm, which ishigh matching error due to pixel-wise transmission. These two methods are eight-path belief propagation arid stereo matching based on color segmentation. Theeight-path BP method improves the accuracy by involving more pixels intoinformation transmission. Meanwhile, deeper research into local disparityoptimization base on segmentation has been done in this paper，which combinesthe Mean-Shitf method and segment the initial image by the color information. Thecolor consistency area will be extracted, then optimize the initial match result withsimilarity measure function to obtain more accurate disparity map. Comparingwith other algorithms, the method in this paper is better. There,re also theshortcomings of the algorithm in this thesis and the problems that need furtherstudy.